Bug Fix for offload_states API#7050
Merged
tjruwase merged 6 commits intodeepspeedai:masterfrom Feb 21, 2025
Merged
Conversation
bugfix for offload_states Signed-off-by: Wei Wu <wuwei211x@gmail.com>
U-rara
force-pushed
the
bugfix_reload_states
branch
from
00fa6a4 to
836b55b
Compare
tjruwase
reviewed
Feb 20, 2025
tjruwase
reviewed
Feb 20, 2025
Signed-off-by: Wei Wu <wuwei211x@gmail.com>
Contributor
Author
|
@tjruwase Good suggestion! My previous fix seemed a bit hasty, so I optimized it according to your advice. |
tjruwase
approved these changes
Feb 20, 2025
github-merge-queue
Bot
removed this pull request from the merge queue due to no response for status checks
Yejing-Lai
pushed a commit
to Yejing-Lai/DeepSpeed
that referenced
this pull request
Feb 24, 2025
@fukun07 and I discovered a bug when using the `offload_states` and `reload_states` APIs of the Zero3 optimizer. When using grouped parameters (for example, in weight decay or grouped lr scenarios), the order of the parameters mapping in `reload_states` ([here](https://github.com/deepspeedai/DeepSpeed/blob/14b3cce4aaedac69120d386953e2b4cae8c2cf2c/deepspeed/runtime/zero/stage3.py#L2953)) does not correspond with the initialization of `self.lp_param_buffer` ([here](https://github.com/deepspeedai/DeepSpeed/blob/14b3cce4aaedac69120d386953e2b4cae8c2cf2c/deepspeed/runtime/zero/stage3.py#L731)), which leads to misaligned parameter loading. This issue was overlooked by the corresponding unit tests ([here](https://github.com/deepspeedai/DeepSpeed/blob/master/tests/unit/runtime/zero/test_offload_states.py)), so we fixed the bug in our PR and added the corresponding unit tests. --------- Signed-off-by: Wei Wu <wuwei211x@gmail.com> Co-authored-by: Masahiro Tanaka <81312776+tohtana@users.noreply.github.com>
deepcharm
pushed a commit
to deepcharm/DeepSpeed
that referenced
this pull request
Feb 26, 2025
@fukun07 and I discovered a bug when using the `offload_states` and `reload_states` APIs of the Zero3 optimizer. When using grouped parameters (for example, in weight decay or grouped lr scenarios), the order of the parameters mapping in `reload_states` ([here](https://github.com/deepspeedai/DeepSpeed/blob/14b3cce4aaedac69120d386953e2b4cae8c2cf2c/deepspeed/runtime/zero/stage3.py#L2953)) does not correspond with the initialization of `self.lp_param_buffer` ([here](https://github.com/deepspeedai/DeepSpeed/blob/14b3cce4aaedac69120d386953e2b4cae8c2cf2c/deepspeed/runtime/zero/stage3.py#L731)), which leads to misaligned parameter loading. This issue was overlooked by the corresponding unit tests ([here](https://github.com/deepspeedai/DeepSpeed/blob/master/tests/unit/runtime/zero/test_offload_states.py)), so we fixed the bug in our PR and added the corresponding unit tests. --------- Signed-off-by: Wei Wu <wuwei211x@gmail.com> Co-authored-by: Masahiro Tanaka <81312776+tohtana@users.noreply.github.com> Signed-off-by: Max Kovalenko <mkovalenko@habana.ai>
deepcharm
pushed a commit
to deepcharm/DeepSpeed
that referenced
this pull request
Feb 27, 2025
@fukun07 and I discovered a bug when using the `offload_states` and `reload_states` APIs of the Zero3 optimizer. When using grouped parameters (for example, in weight decay or grouped lr scenarios), the order of the parameters mapping in `reload_states` ([here](https://github.com/deepspeedai/DeepSpeed/blob/14b3cce4aaedac69120d386953e2b4cae8c2cf2c/deepspeed/runtime/zero/stage3.py#L2953)) does not correspond with the initialization of `self.lp_param_buffer` ([here](https://github.com/deepspeedai/DeepSpeed/blob/14b3cce4aaedac69120d386953e2b4cae8c2cf2c/deepspeed/runtime/zero/stage3.py#L731)), which leads to misaligned parameter loading. This issue was overlooked by the corresponding unit tests ([here](https://github.com/deepspeedai/DeepSpeed/blob/master/tests/unit/runtime/zero/test_offload_states.py)), so we fixed the bug in our PR and added the corresponding unit tests. --------- Signed-off-by: Wei Wu <wuwei211x@gmail.com> Co-authored-by: Masahiro Tanaka <81312776+tohtana@users.noreply.github.com> Signed-off-by: Max Kovalenko <mkovalenko@habana.ai>
gyou2021
pushed a commit
to gyou2021/DeepSpeed
that referenced
this pull request
Feb 28, 2025
@fukun07 and I discovered a bug when using the `offload_states` and `reload_states` APIs of the Zero3 optimizer. When using grouped parameters (for example, in weight decay or grouped lr scenarios), the order of the parameters mapping in `reload_states` ([here](https://github.com/deepspeedai/DeepSpeed/blob/14b3cce4aaedac69120d386953e2b4cae8c2cf2c/deepspeed/runtime/zero/stage3.py#L2953)) does not correspond with the initialization of `self.lp_param_buffer` ([here](https://github.com/deepspeedai/DeepSpeed/blob/14b3cce4aaedac69120d386953e2b4cae8c2cf2c/deepspeed/runtime/zero/stage3.py#L731)), which leads to misaligned parameter loading. This issue was overlooked by the corresponding unit tests ([here](https://github.com/deepspeedai/DeepSpeed/blob/master/tests/unit/runtime/zero/test_offload_states.py)), so we fixed the bug in our PR and added the corresponding unit tests. --------- Signed-off-by: Wei Wu <wuwei211x@gmail.com> Co-authored-by: Masahiro Tanaka <81312776+tohtana@users.noreply.github.com> Signed-off-by: gyou2021 <ganmei.you@intel.com>
tohtana
added a commit
that referenced
this pull request
Feb 28, 2025
@fukun07 and I discovered a bug when using the `offload_states` and `reload_states` APIs of the Zero3 optimizer. When using grouped parameters (for example, in weight decay or grouped lr scenarios), the order of the parameters mapping in `reload_states` ([here](https://github.com/deepspeedai/DeepSpeed/blob/14b3cce4aaedac69120d386953e2b4cae8c2cf2c/deepspeed/runtime/zero/stage3.py#L2953)) does not correspond with the initialization of `self.lp_param_buffer` ([here](https://github.com/deepspeedai/DeepSpeed/blob/14b3cce4aaedac69120d386953e2b4cae8c2cf2c/deepspeed/runtime/zero/stage3.py#L731)), which leads to misaligned parameter loading. This issue was overlooked by the corresponding unit tests ([here](https://github.com/deepspeedai/DeepSpeed/blob/master/tests/unit/runtime/zero/test_offload_states.py)), so we fixed the bug in our PR and added the corresponding unit tests. --------- Signed-off-by: Wei Wu <wuwei211x@gmail.com> Co-authored-by: Masahiro Tanaka <81312776+tohtana@users.noreply.github.com> Signed-off-by: Masahiro Tanaka <mtanaka@microsoft.com>
shenzheyu
pushed a commit
to shenzheyu/DeepSpeed
that referenced
this pull request
Mar 5, 2025
@fukun07 and I discovered a bug when using the `offload_states` and `reload_states` APIs of the Zero3 optimizer. When using grouped parameters (for example, in weight decay or grouped lr scenarios), the order of the parameters mapping in `reload_states` ([here](https://github.com/deepspeedai/DeepSpeed/blob/14b3cce4aaedac69120d386953e2b4cae8c2cf2c/deepspeed/runtime/zero/stage3.py#L2953)) does not correspond with the initialization of `self.lp_param_buffer` ([here](https://github.com/deepspeedai/DeepSpeed/blob/14b3cce4aaedac69120d386953e2b4cae8c2cf2c/deepspeed/runtime/zero/stage3.py#L731)), which leads to misaligned parameter loading. This issue was overlooked by the corresponding unit tests ([here](https://github.com/deepspeedai/DeepSpeed/blob/master/tests/unit/runtime/zero/test_offload_states.py)), so we fixed the bug in our PR and added the corresponding unit tests. --------- Signed-off-by: Wei Wu <wuwei211x@gmail.com> Co-authored-by: Masahiro Tanaka <81312776+tohtana@users.noreply.github.com> Signed-off-by: Zheyu SHEN <zyshen@umd.edu>
ys950902
pushed a commit
to ys950902/DeepSpeed
that referenced
this pull request
Mar 6, 2025
@fukun07 and I discovered a bug when using the `offload_states` and `reload_states` APIs of the Zero3 optimizer. When using grouped parameters (for example, in weight decay or grouped lr scenarios), the order of the parameters mapping in `reload_states` ([here](https://github.com/deepspeedai/DeepSpeed/blob/14b3cce4aaedac69120d386953e2b4cae8c2cf2c/deepspeed/runtime/zero/stage3.py#L2953)) does not correspond with the initialization of `self.lp_param_buffer` ([here](https://github.com/deepspeedai/DeepSpeed/blob/14b3cce4aaedac69120d386953e2b4cae8c2cf2c/deepspeed/runtime/zero/stage3.py#L731)), which leads to misaligned parameter loading. This issue was overlooked by the corresponding unit tests ([here](https://github.com/deepspeedai/DeepSpeed/blob/master/tests/unit/runtime/zero/test_offload_states.py)), so we fixed the bug in our PR and added the corresponding unit tests. --------- Signed-off-by: Wei Wu <wuwei211x@gmail.com> Co-authored-by: Masahiro Tanaka <81312776+tohtana@users.noreply.github.com> Signed-off-by: yisheng <yi.sheng@intel.com>
mauryaavinash95
pushed a commit
to DataStates/DeepSpeed
that referenced
this pull request
Mar 20, 2025
@fukun07 and I discovered a bug when using the `offload_states` and `reload_states` APIs of the Zero3 optimizer. When using grouped parameters (for example, in weight decay or grouped lr scenarios), the order of the parameters mapping in `reload_states` ([here](https://github.com/deepspeedai/DeepSpeed/blob/14b3cce4aaedac69120d386953e2b4cae8c2cf2c/deepspeed/runtime/zero/stage3.py#L2953)) does not correspond with the initialization of `self.lp_param_buffer` ([here](https://github.com/deepspeedai/DeepSpeed/blob/14b3cce4aaedac69120d386953e2b4cae8c2cf2c/deepspeed/runtime/zero/stage3.py#L731)), which leads to misaligned parameter loading. This issue was overlooked by the corresponding unit tests ([here](https://github.com/deepspeedai/DeepSpeed/blob/master/tests/unit/runtime/zero/test_offload_states.py)), so we fixed the bug in our PR and added the corresponding unit tests. --------- Signed-off-by: Wei Wu <wuwei211x@gmail.com> Co-authored-by: Masahiro Tanaka <81312776+tohtana@users.noreply.github.com>
loadams
pushed a commit
that referenced
this pull request
Mar 25, 2025
@fukun07 and I discovered a bug when using the `offload_states` and `reload_states` APIs of the Zero3 optimizer. When using grouped parameters (for example, in weight decay or grouped lr scenarios), the order of the parameters mapping in `reload_states` ([here](https://github.com/deepspeedai/DeepSpeed/blob/14b3cce4aaedac69120d386953e2b4cae8c2cf2c/deepspeed/runtime/zero/stage3.py#L2953)) does not correspond with the initialization of `self.lp_param_buffer` ([here](https://github.com/deepspeedai/DeepSpeed/blob/14b3cce4aaedac69120d386953e2b4cae8c2cf2c/deepspeed/runtime/zero/stage3.py#L731)), which leads to misaligned parameter loading. This issue was overlooked by the corresponding unit tests ([here](https://github.com/deepspeedai/DeepSpeed/blob/master/tests/unit/runtime/zero/test_offload_states.py)), so we fixed the bug in our PR and added the corresponding unit tests. --------- Signed-off-by: Wei Wu <wuwei211x@gmail.com> Co-authored-by: Masahiro Tanaka <81312776+tohtana@users.noreply.github.com> Signed-off-by: Logan Adams <loadams@microsoft.com>
hijkzzz
pushed a commit
to OpenRLHF/OpenRLHF
that referenced
this pull request
Apr 8, 2025
Those offload types are fixed in deepspeedai/DeepSpeed#7050, and they have already been released in 0.16.5. Signed-off-by: Hollow Man <hollowman@opensuse.org>
LXXXXR
pushed a commit
to LXXXXR/AESL
that referenced
this pull request
Feb 15, 2026
Those offload types are fixed in deepspeedai/DeepSpeed#7050, and they have already been released in 0.16.5. Signed-off-by: Hollow Man <hollowman@opensuse.org>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
@fukun07 and I discovered a bug when using the
offload_statesandreload_statesAPIs of the Zero3 optimizer. When using grouped parameters (for example, in weight decay or grouped lr scenarios), the order of the parameters mapping inreload_states(here) does not correspond with the initialization ofself.lp_param_buffer(here), which leads to misaligned parameter loading. This issue was overlooked by the corresponding unit tests (here), so we fixed the bug in our PR and added the corresponding unit tests.